STL-Robust Mixture-of-Experts Gating / bench_report.json

Unverified

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.0047177650849334896
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0023727577936369926
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.0009645432000979781
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0009824210283113644,
 33      "std": 0.0002772167467598855,
 34      "per_seed": [
 35        0.0006978703895583749,
 36        0.0009150496916845441,
 37        0.0006460942095145583,
 38        0.0015991585096344352,
 39        0.000958278018515557,
 40        0.0008829828002490103,
 41        0.0010238300310447812,
 42        0.0011361045762896538
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.0005874872585991398,
 49    "std": 0.00012433054166616667,
 50    "per_seed": [
 51      0.0004255748353898525,
 52      0.0006285446579568088,
 53      0.0005254852003417909,
 54      0.0007636661757715046,
 55      0.0006406739703379571,
 56      0.00038608635077252984,
 57      0.0006125259678810835,
 58      0.0007173409103415906
 59    ],
 60    "n": 8,
 61    "selected_cfg": {
 62      "lr": 0.006,
 63      "beta": 1.5
 64    }
 65  },
 66  "comparison": {
 67    "delta_mean": -0.0003949337697122246,
 68    "idea_wins": 8,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      -0.00027229555416852236,
 72      -0.0002865050337277353,
 73      -0.0001206090091727674,
 74      -0.0008354923338629305,
 75      -0.0003176040481775999,
 76      -0.0004968964494764805,
 77      -0.0004113040631636977,
 78      -0.00041876366594806314
 79    ],
 80    "p_value": 0.0081,
 81    "mde": 0.00017691123938093242,
 82    "mde_rel_pct": 18.007680442774777,
 83    "verdict": "idea better (significant)",
 84    "system_worked": true
 85  },
 86  "mechanism_signature": {
 87    "track_choice": "dynamics: pendulum control and long-term stability structure",
 88    "math_check": {
 89      "predicted_slope": 1.3,
 90      "observed_slope": 1.2999999999999996,
 91      "max_identity_error": 8.881784197001252e-16
 92    },
 93    "idea_sweep": [
 94      {
 95        "cfg": {
 96          "lr": 0.006,
 97          "beta": 0.0
 98        },
 99        "mean": 0.0009645432000979781
100      },
101      {
102        "cfg": {
103          "lr": 0.006,
104          "beta": 0.75
105        },
106        "mean": 0.0009360480034956709
107      },
108      {
109        "cfg": {
110          "lr": 0.006,
111          "beta": 1.5
112        },
113        "mean": 0.0005858177173649892
114      }
115    ],
116    "mechanism_signature": {
117      "beta": 1.5,
118      "predicted_slope": 1.5,
119      "observed_slope": 1.4858029248869011,
120      "robustness_logodds_correlation": 0.9999296076539561,
121      "n_samples": 256,
122      "confirmed": true
123    }
124  }
125}